ABSTRACT

Painting is one of the popular forms of art. To redraw a picture in a certain style in the past required a skilled artist and a lot of work. There are many techniques and studies researching how to turn pictures into beautiful art works. Among these studies, the convolutional neural network is one of the deep learning techniques used in creating artistic images by dividing up and recombining the content and style of the images. This image editing process is called Neural Style Transfer (NST). It’s an optimization approach that combines two images known as a content image and a style reference image (say, an artwork from a painter) – so that the output image resembles the content image, but appears to have been painted in a style reference image way. For images classification, Visual Geometry Group (VGG) is most popular algorithm because of its pre-trained with several classes and huge amount of data. The style will be applied to the content image effectively using this model. To get the best art image by computing the high-level features in content image and low-level features along with gram matrices in style image. Optimization algorithm such as Limited-memory BFGS is used to optimize the resultant image. Ultimately, the Artistic neural style transfer gives the AI painted image.

Keywords: VGG, Neural Style Transfer, Transfer learning, Convolutional neural network, Image classification.